I’ve been hunting around for a good, brief explanation for you, but can’t find one.

Generally, explaining how population weighting works with a decent level of detail takes up about a whole textbook (I’ve got three textbooks on this, and they’re all about 300-400 pages long). There’s an enormous amount of detail involved since it’s one of the most important things a pollster does.

At it’s most basic, the raw respondent results are collated and compared to known levels of the population for certain attributes like age, sex, geography etc- so we know that the whole voting population has so many people between the ages of X and Y, and another proportion of people between Y and Z. The individual responses in the sample are then weighted by those known proportions so that the sample itself more closely approximates the true voting population.

The complexity comes in the numerous ways that this can be achieved – because effectively there’s more than one way to skin this particular statistical cat.

In theory it does blow out the sampling error, but the gains in accuracy achieved by this sort of post-stratification weighting are greater than the losses due to the increase in theoretical sampling error. It’s actually possible to test the cost/benefit of a post-stratification weighting system when it comes to accuracy after every election result. (assuming that the population didnt change their mind on who to vote for between the time of the final poll of the election campaign and election day itself).

I’ll keep an eye out if I can find a relatively brief but still detailed methodology for how this works, and I’ll let you know if I come across one.

...] the significant socieconomic factors which influence student results. An Essential Research survey, reported at Pollytics blog, starts to explore this [...

...] latest tax poll from Essential Research, published on Pollytics blog, shows again that question wording effects make opinions on tax and spend hard to [...

Anyhow it seems ok now, even before I try GG’s trick, and by luck my bro in law, who is a computer pro, is visiting and I’ll get him onto it tomorrow.

The site content is 600 pixels wide, the sidebar is 300 pixels wide, so a person really needs a minimum 900 pixel wide resolution for everything to work properly.

That seems to be the most common issue here, the width, but I’ve never heard of the comment box obscuring stuff upwards.

What’s your browser and screen resolution?

BTW Poss frequently when I arrive here the comments box obscures the last response, so currently I can’t read what OzPol has written.
I presume this is a Crikey site whatever ‘feature’ unless its a feature of my computer set up, do other people get it?

My concern is that presumably the results get scaled based on some arbitrary demographic cut, e.g. our survey had X% males between the age of 18-25 but the polulation has Y% of this group so the responses for this group are weighted by Y/X. Wouldn’t this introduce additional sampling error, since the number of responses for each group would presumably be quite small, and then the error gets multiplied…

As I say, an explanation of how this works would be handy.